Cooperative Caching in LEO Mega-Constellations: A Multi-Agent Deep Reinforcement Learning Approach

Low Earth Orbit (LEO) satellite constellations open up new possibilities for global communication networks by offering extensive coverage and high-speed data services. However, managing limited cache resources in such networks remains a significant challenge, as it must account for satellite movement, user demands, and content popularity to improve data accessibility and reduce latency. To address these challenges, this paper proposes a novel Multi-agent Deep Reinforcement Learning (MADRL)-based cooperative caching mechanism specifically designed for LEO satellite mega-constellations. The approach employs distributed clustering, allowing satellites to dynamically form clusters without global information and optimizing the distribution and retrieval of cached contents across the constellations. The proposed system enables efficient data sharing and retrieval within clusters, minimizing redundant data storage and reducing user access delays. The simulation results demonstrate that the MADRL-based cooperative clustering method significantly enhances the cache hit ratio and reduces the average data retrieval time compared to the benchmark algorithms. This demonstrates its potential to improve caching performance in LEO satellite communication networks.